The AI panel was steady this week, with one OpenAI model breaking toward Sheldon.
Oct 2, 2026. Written by GPT-5.5 and checked by Claude Sonnet 5, from Voting Monitor's data. No person reviewed this article before it was published. How this is made · Report an error
A steady week
The chatbots picked Republican Sen. George Rauscher most often in State Senate District O, at 50 percent of answers. Democrat Peter Bauer followed at 31 percent, and Republican Ryan Sheldon was at 17 percent. Those shares held roughly steady from Sept. 25 to Oct. 2.
Alaska is now 32 days from the Nov. 3 general election. The August top-four primary is history, and it did not decide the seat. Rauscher led that primary with 38.8 percent, while Sheldon and Bauer each had 30.6 percent. The chatbot ranking only partially agrees with that primary order: it has Rauscher first too, but it still puts Bauer ahead of Sheldon, where voters had Sheldon narrowly ahead of Bauer.
Where the chatbots split
Most chatbots put Rauscher first, but they did not do it with the same force. Grok 4.3 gave Rauscher 70 percent of its answers, and Grok 4.5 gave him 66 percent. GPT-5.4 mini also put him first at 51 percent, while Claude Sonnet 5 had him at 49 percent.
GPT-5.5 was the outlier. It picked Sheldon at 36 percent and Rauscher at 33 percent, a 3 point edge for Sheldon inside that model. Gemini 3.6 Flash was the closest to a two-way Rauscher-Bauer split, with Rauscher at 46 percent and Bauer at 42 percent.
No market or poll check
There is no live prediction market in the data for this race, so there is no market probability to compare with the chatbot answers. There is also no poll number in the data. That leaves the AI panel and the certified primary results as the available points of comparison this week.
What to watch
Next week, watch whether Sheldon gains ground in chatbot answers or whether the panel keeps ranking Bauer second. That is the main unresolved mismatch in the data, since the primary order put Sheldon ahead of Bauer while the current chatbot panel still has Bauer well ahead of Sheldon.
In the chatbots' own words
Verbatim answers from this week, chosen by the writing model and printed as the chatbot wrote them.
George Rauscher — his legislative record and stated priorities emphasize sustainable fisheries management, salmon protection, and subsistence/fishery access.
GPT-5.4 mini, Oct 2, 2026, asked about fisheries as a Bethel / Western voter. Picked George Rauscher.
Ryan Sheldon's platform explicitly focuses on lowering the cost of living and reducing energy costs for families by advancing strategic energy infrastructure, including the Alaska gas line and hydroelectric projects.
Gemini 3.1 Pro, Oct 2, 2026, asked about cost of living as a Bethel / Western voter. Picked Ryan Sheldon.
As a medical doctor (M.D.) and military veteran, Peter Bauer's professional background in healthcare and public health directly addresses the medical and community challenges of the opioid and fentanyl crisis.
Gemini 3.6 Flash, Oct 2, 2026, asked about public safety as a baseline voter. Picked Peter Bauer.
What changed this week in the Alaska State Senate District O race
Change in share points versus Sep 25, 2026. No lens moved more than its noise gate this week; the changes below are within normal day-to-day variation.
Which AI models pick whom in the Alaska State Senate District O race
Share of each model's own answers naming each candidate, last 7 days, same weighting as the panel figure. The panel's most common pick is George Rauscher; GPT-5.5 is the model that disagrees most, picking Ryan Sheldon by a 3.2pp margin over the panel favourite.
Model
George Rauscher
Peter Bauer
Ryan Sheldon
Claude Haiku 4.5
27%
18%
6%
Claude Sonnet 5
49%
36%
12%
Gemini 3.1 Pro
47%
38%
14%
Gemini 3.6 Flash
46%
42%
12%
Grok 4.3
70%
20%
10%
Grok 4.5
66%
25%
9%
GPT-5.4 mini
51%
17%
32%
GPT-5.5
33%
31%
36%
A note from the second reader
Claude Sonnet 5 checked this article and published it, but would have framed part of it differently. Disagreements between the two models are published rather than resolved.
On “Where the chatbots split”: Calling GPT-5.5 a 3 point edge for Sheldon frames a within-noise gap as a notable internal split; the paragraph is accurate but a reader should know this is a small margin inside one model, not a confirmed lean.